“are llms generative ai”

Based on the search results, here is a summary on whether LLMs (large language models) can be considered generative AI:

LLMs can be considered generative models, but with some caveats:

• LLMs are trained to predict the next token in a sequence, which can be seen as a form of generation . The output of an LLM is a probability distribution over the next possible tokens.

• Technically, as long as an AI model can generate samples similar to the input distribution, it can be considered a generative model . LLMs meet this criteria by generating plausible text.

• However, some argue that LLMs are not truly "generative" in the sense of being able to generate novel, creative content from scratch. They are more akin to "glorified autocomplete" .

• LLMs are limited in their reasoning abilities and can produce untrustworthy or hallucinated outputs, which makes them less than completely "trustworthy" generative models .

• An alternative approach proposed is to build AI systems educated with curated knowledge and rules, which could produce more trustworthy and interpretable outputs, though at the cost of slower inference speed .

In summary, while LLMs exhibit some generative capabilities, there are open questions and debates around whether they can be considered true generative AI models given their limitations in reasoning and trustworthiness. The field is still evolving.